Summary of project + methods

This study is examining the association between DNA methylation and childhood eczema (up to ~10 years). Studies have data from different timepoints, so in the first instance we will classify individuals as ‘any eczema’ versus ‘no eczema’ according to all timepoints available up to age 10. Sub-analyses will be conducted that limits cases to those children who are diagnosed early in childhood (by ~2 years) or have persistent eczema (diagnosed by age 2 and have current eczema at age 8yrs).

The analyses will therefore use 3 binary definitions:

  • Childhood AD - cases: AD by age 10, controls:no AD by age 10
  • Early-onset AD - cases: AD by age 2, controls: no AD by age 2
  • Persistent AD - cases: AD by age 2 and persisting to ~ 8-10yrs, controls: no AD by age 10

The association of each outcome with DNA methylation will be explored using logistic regression and 3 models with different covariates:

  • model a: sex + batch (+ selection_factors) (+ ancestry)
  • model b: sex + batch + maternal_age + maternal_smoking + maternal_SES + gestational_age (+selection_factors) (+ ancestry)
  • model c: sex + batch + maternal_age + maternal_smoking + maternal_SES + gestational_age + cell_type (+ selection_factors) (+ ancestry)

In this report, the correlation between cohort effect estimates is examined. Further the relationship between effect estimate correlation and difference between prevalence estimates is assessed.

After the summary below and under each eczema type and model there are tables that show pairwise comparisons of each cohort. The correlation of effect estimates across the top 30 CpGs (the 30 CpGs with the lowest P values in the meta-analysis of EWAS) are shown (under the “effect_cor” column heading) alongside the difference in prevalence estimates between the cohorts (under the “prev_diff” column heading).

Using the top 30 CpG sites from the meta-analysis for each model, M-statistics were also calculated. These statistics give an indication of the study-wide heterogeneity across CpG sites, rather than typical heterogeneity statistics (such as I2) that assess heterogeneity across studies for single CpG sites.

Summary of models and association between effect estimate correlation and prevalence difference

Table 1: A summary of the association between effect size correlations and prevalence differences between cohorts
eczema-definition model n-tophits maxp top-hit-wilc-p all-wilc-p
childhood a 30 7.7e-05 1.8e-14 1.7e-14
childhood b 30 5.7e-05 1.7e-14 1.7e-14
childhood c 30 1.8e-04 1.9e-12 1.7e-12
early-onset a 30 5.7e-05 1.4e-16 1.2e-16
early-onset b 30 7.9e-05 1.4e-16 1.2e-16
early-onset c 30 7.4e-05 1.4e-16 1.2e-16
persistent a 30 7.9e-05 5.8e-11 2.9e-11
persistent b 30 7.5e-05 5.8e-11 2.9e-11
persistent c 30 7.9e-05 2.9e-11 2.9e-11
Correlations between effect sizes from different cohorts were calculated and then the association between these correlations and prevalence differences between the cohorts was estimated using a paired Wilcoxon signed rank test. This was analysis was performed using just the top 30 CpG sites from the meta-analysis (by P value) or using all CpG sites that overlapped between the individual cohort EWAS. n-tophits = number of CpGs selected from the meta-analysis, maxp = highest P value from the top CpGs, top-hit-wilc-p = P value from the Wilcoxon signed rank test when using the top CpGs only, all-wilc-p = P value from the Wilcoxon signed rank test when using all CpGs.

Childhood AD

Sample sizes + prevalence

cohort N N-cases N-controls prevalence
ALSPAC 673 343 330 51.0
CHS 197 50 147 25.4
EDEN 153 69 84 45.1
GENR 450 69 381 15.3
INMA 462 216 246 46.8
MoBa1 831 399 432 48.0
MoBa2 418 179 239 42.8
NEST_B 77 56 21 72.7
NEST_W 41 22 19 53.7
PREDO 792 54 738 6.8
VITO 78 23 55 29.5
GOYA 517 93 424 18.0
IOW 616 179 437 29.1
Total 5,305 1,752 3,553 33.0

Meta-regression results

Table 2: Assessing the impact of prevalence and samplesize on the influence a study has in the meta-analysis results
model variable beta se zval pval lower-ci upper-ci
m1a N -0.00051 0.00024 -2.12 0.034 -0.00098 -3.9e-05
m1a N_cases -0.00031 0.00064 -0.49 0.628 -0.00157 9.5e-04
m1a prevalence 0.00824 0.00352 2.34 0.019 0.00134 1.5e-02
m1a definition 0.15479 0.15367 1.01 0.314 -0.14639 4.6e-01
m1a definition_3grp 0.05426 0.15402 0.35 0.725 -0.24761 3.6e-01
m1b N -0.00043 0.00029 -1.49 0.136 -0.00099 1.3e-04
m1b N_cases -0.00032 0.00072 -0.44 0.656 -0.00173 1.1e-03
m1b prevalence 0.00745 0.00415 1.79 0.073 -0.00069 1.6e-02
m1b definition 0.14632 0.20392 0.72 0.473 -0.25335 5.5e-01
m1b definition_3grp 0.03536 0.21558 0.16 0.870 -0.38717 4.6e-01
m1c N -0.00004 0.00020 -0.20 0.844 -0.00044 3.6e-04
m1c N_cases 0.00031 0.00044 0.70 0.482 -0.00055 1.2e-03
m1c prevalence 0.00223 0.00284 0.79 0.432 -0.00333 7.8e-03
m1c definition 0.24785 0.12568 1.97 0.049 0.00153 4.9e-01
m1c definition_3grp 0.16596 0.12615 1.32 0.188 -0.08129 4.1e-01

model a

Table 3: childhood AD, model a pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.724 25.59
ALSPAC EDEN 0.518 5.87
CHS EDEN 0.325 19.72
ALSPAC GENR 0.700 35.63
CHS GENR 0.506 10.05
EDEN GENR 0.698 29.76
ALSPAC GOYA 0.074 32.98
CHS GOYA -0.241 7.39
EDEN GOYA -0.147 27.11
GENR GOYA -0.157 2.66
ALSPAC INMA 0.687 4.21
CHS INMA 0.567 21.37
EDEN INMA 0.597 1.66
GENR INMA 0.779 31.42
GOYA INMA -0.078 28.76
ALSPAC IOW 0.761 21.91
CHS IOW 0.564 3.68
EDEN IOW 0.575 16.04
GENR IOW 0.753 13.73
GOYA IOW -0.153 11.07
INMA IOW 0.855 17.69
ALSPAC Meta 0.914 NA
CHS Meta 0.711 NA
EDEN Meta 0.634 NA
GENR Meta 0.843 NA
GOYA Meta -0.013 NA
INMA Meta 0.880 NA
IOW Meta 0.908 NA
ALSPAC MoBa1 0.708 2.95
CHS MoBa1 0.509 22.63
EDEN MoBa1 0.601 2.92
GENR MoBa1 0.765 32.68
GOYA MoBa1 -0.062 30.03
INMA MoBa1 0.803 1.26
IOW MoBa1 0.837 18.96
Meta MoBa1 0.865 NA
ALSPAC MoBa2 0.407 8.14
CHS MoBa2 0.258 17.44
EDEN MoBa2 0.482 2.28
GENR MoBa2 0.490 27.49
GOYA MoBa2 0.131 24.83
INMA MoBa2 0.253 3.93
IOW MoBa2 0.298 13.76
Meta MoBa2 0.460 NA
MoBa1 MoBa2 0.284 5.19
ALSPAC NEST_B 0.714 21.76
CHS NEST_B 0.551 47.35
EDEN NEST_B 0.677 27.63
GENR NEST_B 0.713 57.39
GOYA NEST_B -0.121 54.74
INMA NEST_B 0.831 25.97
IOW NEST_B 0.833 43.67
Meta NEST_B 0.845 NA
MoBa1 NEST_B 0.809 24.71
MoBa2 NEST_B 0.201 29.90
ALSPAC NEST_W 0.041 2.69
CHS NEST_W -0.315 28.28
EDEN NEST_W 0.421 8.56
GENR NEST_W 0.396 38.33
GOYA NEST_W 0.241 35.67
INMA NEST_W 0.417 6.91
IOW NEST_W 0.328 24.60
Meta NEST_W 0.260 NA
MoBa1 NEST_W 0.373 5.64
MoBa2 NEST_W 0.204 10.84
NEST_B NEST_W 0.336 19.07
ALSPAC PREDO 0.793 44.15
CHS PREDO 0.622 18.56
EDEN PREDO 0.578 38.28
GENR PREDO 0.786 8.52
GOYA PREDO 0.083 11.17
INMA PREDO 0.713 39.94
IOW PREDO 0.729 22.24
Meta PREDO 0.865 NA
MoBa1 PREDO 0.628 41.20
MoBa2 PREDO 0.562 36.00
NEST_B PREDO 0.681 65.91
NEST_W PREDO 0.248 46.84
ALSPAC VITO 0.580 21.48
CHS VITO 0.587 4.11
EDEN VITO 0.121 15.61
GENR VITO 0.505 14.15
GOYA VITO -0.075 11.50
INMA VITO 0.549 17.27
IOW VITO 0.500 0.43
Meta VITO 0.607 NA
MoBa1 VITO 0.404 18.53
MoBa2 VITO 0.176 13.34
NEST_B VITO 0.459 43.24
NEST_W VITO -0.104 24.17
PREDO VITO 0.491 22.67
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of childhood eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of childhood eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 1: Correlation between the effect estimates of childhood eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 2: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 3: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model b

Table 4: childhood AD, model b pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.868 25.59
ALSPAC EDEN 0.452 5.87
CHS EDEN 0.256 19.72
ALSPAC GENR 0.783 35.63
CHS GENR 0.706 10.05
EDEN GENR 0.559 29.76
ALSPAC GOYA 0.504 32.98
CHS GOYA 0.392 7.39
EDEN GOYA 0.126 27.11
GENR GOYA 0.243 2.66
ALSPAC INMA 0.476 3.60
CHS INMA 0.466 21.99
EDEN INMA 0.412 2.27
GENR INMA 0.371 32.04
GOYA INMA 0.311 29.38
ALSPAC IOW 0.659 21.71
CHS IOW 0.626 3.88
EDEN IOW 0.487 15.84
GENR IOW 0.594 13.92
GOYA IOW 0.205 11.27
INMA IOW 0.601 18.11
ALSPAC Meta 0.936 NA
CHS Meta 0.862 NA
EDEN Meta 0.564 NA
GENR Meta 0.811 NA
GOYA Meta 0.462 NA
INMA Meta 0.675 NA
IOW Meta 0.784 NA
ALSPAC MoBa1 0.745 2.90
CHS MoBa1 0.752 22.69
EDEN MoBa1 0.394 2.97
GENR MoBa1 0.619 32.73
GOYA MoBa1 0.188 30.08
INMA MoBa1 0.494 0.70
IOW MoBa1 0.623 18.81
Meta MoBa1 0.807 NA
ALSPAC MoBa2 0.726 8.14
CHS MoBa2 0.572 17.44
EDEN MoBa2 0.727 2.28
GENR MoBa2 0.757 27.49
GOYA MoBa2 0.301 24.83
INMA MoBa2 0.639 4.55
IOW MoBa2 0.695 13.57
Meta MoBa2 0.849 NA
MoBa1 MoBa2 0.653 5.24
ALSPAC NEST_B 0.518 21.76
CHS NEST_B 0.497 47.35
EDEN NEST_B 0.482 27.63
GENR NEST_B 0.442 57.39
GOYA NEST_B 0.253 54.74
INMA NEST_B 0.512 25.36
IOW NEST_B 0.549 43.47
Meta NEST_B 0.640 NA
MoBa1 NEST_B 0.500 24.66
MoBa2 NEST_B 0.494 29.90
ALSPAC NEST_W 0.078 2.69
CHS NEST_W 0.031 28.28
EDEN NEST_W 0.508 8.56
GENR NEST_W 0.134 38.33
GOYA NEST_W 0.067 35.67
INMA NEST_W 0.378 6.29
IOW NEST_W 0.197 24.40
Meta NEST_W 0.206 NA
MoBa1 NEST_W 0.024 5.59
MoBa2 NEST_W 0.396 10.84
NEST_B NEST_W 0.237 19.07
ALSPAC PREDO 0.833 44.15
CHS PREDO 0.775 18.56
EDEN PREDO 0.488 38.28
GENR PREDO 0.759 8.52
GOYA PREDO 0.383 11.17
INMA PREDO 0.509 40.55
IOW PREDO 0.673 22.44
Meta PREDO 0.883 NA
MoBa1 PREDO 0.600 41.25
MoBa2 PREDO 0.726 36.00
NEST_B PREDO 0.596 65.91
NEST_W PREDO 0.215 46.84
ALSPAC VITO 0.592 21.48
CHS VITO 0.484 4.11
EDEN VITO 0.117 15.61
GENR VITO 0.598 14.15
GOYA VITO 0.523 11.50
INMA VITO 0.146 17.88
IOW VITO 0.216 0.23
Meta VITO 0.523 NA
MoBa1 VITO 0.291 18.58
MoBa2 VITO 0.433 13.34
NEST_B VITO 0.173 43.24
NEST_W VITO -0.127 24.17
PREDO VITO 0.452 22.67
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of childhood eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of childhood eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 4: Correlation between the effect estimates of childhood eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 5: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 6: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model c

Table 5: childhood AD, model c pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.62715 25.59
ALSPAC EDEN 0.35254 5.87
CHS EDEN 0.42439 19.72
ALSPAC GENR 0.40856 35.63
CHS GENR 0.51558 10.05
EDEN GENR 0.55747 29.76
ALSPAC GOYA 0.63175 32.98
CHS GOYA 0.29616 7.39
EDEN GOYA 0.18751 27.11
GENR GOYA 0.21614 2.66
ALSPAC INMA 0.57363 3.60
CHS INMA 0.57320 21.99
EDEN INMA 0.48464 2.27
GENR INMA 0.40610 32.04
GOYA INMA 0.37862 29.38
ALSPAC IOW 0.71055 21.71
CHS IOW 0.70628 3.88
EDEN IOW 0.36014 15.84
GENR IOW 0.33691 13.92
GOYA IOW 0.41357 11.27
INMA IOW 0.77705 18.11
ALSPAC Meta 0.87786 NA
CHS Meta 0.76469 NA
EDEN Meta 0.52273 NA
GENR Meta 0.61841 NA
GOYA Meta 0.57536 NA
INMA Meta 0.80060 NA
IOW Meta 0.85849 NA
ALSPAC MoBa1 0.66705 2.90
CHS MoBa1 0.65597 22.69
EDEN MoBa1 0.46307 2.97
GENR MoBa1 0.63112 32.73
GOYA MoBa1 0.42647 30.08
INMA MoBa1 0.79494 0.70
IOW MoBa1 0.70355 18.81
Meta MoBa1 0.87096 NA
ALSPAC MoBa2 0.67930 8.14
CHS MoBa2 0.48053 17.44
EDEN MoBa2 0.61650 2.28
GENR MoBa2 0.72558 27.49
GOYA MoBa2 0.52847 24.83
INMA MoBa2 0.48847 4.55
IOW MoBa2 0.55732 13.57
Meta MoBa2 0.74569 NA
MoBa1 MoBa2 0.60287 5.24
ALSPAC NEST_B 0.03950 21.76
CHS NEST_B -0.04220 47.35
EDEN NEST_B 0.30469 27.63
GENR NEST_B 0.09003 57.39
GOYA NEST_B -0.00019 54.74
INMA NEST_B -0.05804 25.36
IOW NEST_B -0.22943 43.47
Meta NEST_B 0.01553 NA
MoBa1 NEST_B -0.03350 24.66
MoBa2 NEST_B -0.05051 29.90
ALSPAC PREDO 0.78875 44.15
CHS PREDO 0.73160 18.56
EDEN PREDO 0.35564 38.28
GENR PREDO 0.58113 8.52
GOYA PREDO 0.46070 11.17
INMA PREDO 0.61208 40.55
IOW PREDO 0.69849 22.44
Meta PREDO 0.86363 NA
MoBa1 PREDO 0.69942 41.25
MoBa2 PREDO 0.58478 36.00
NEST_B PREDO 0.15223 65.91
ALSPAC VITO 0.23321 21.48
CHS VITO 0.44731 4.11
EDEN VITO 0.29879 15.61
GENR VITO 0.01956 14.15
GOYA VITO 0.13552 11.50
INMA VITO 0.30730 17.88
IOW VITO 0.33803 0.23
Meta VITO 0.34064 NA
MoBa1 VITO 0.24834 18.58
MoBa2 VITO 0.18651 13.34
NEST_B VITO 0.05068 43.24
PREDO VITO 0.21045 22.67
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of childhood eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of childhood eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 7: Correlation between the effect estimates of childhood eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 8: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 9: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Early-onset AD

Sample sizes + prevalence

cohort N N-cases N-controls prevalence
ALSPAC 667 169 498 25.3
CHS 197 41 156 20.8
DCHS 266 26 240 9.8
EDEN 142 48 94 33.8
GENR 742 171 571 23.0
INMA 348 83 265 23.9
IOW-F2 171 28 143 16.4
MoBa1 993 361 632 36.4
MoBa2 386 210 176 54.4
NEST_B 119 30 89 25.2
UKIDS 607 216 390 35.6
VITO 70 15 55 21.4
GOYA 517 39 478 7.5
IOW 667 99 569 14.8
Total 5,892 1,536 4,356 26.1

Meta-regression results

Table 6: Assessing the impact of prevalence and samplesize on the influence a study has in the meta-analysis results
model variable beta se zval pval lower-ci upper-ci
m2a N -3.3e-04 0.00031 -1.06 0.287 -0.00094 2.8e-04
m2a N_cases -1.2e-03 0.00083 -1.41 0.159 -0.00279 4.6e-04
m2a prevalence -7.9e-03 0.00711 -1.12 0.264 -0.02188 6.0e-03
m2a definition 5.3e-02 0.17708 0.30 0.767 -0.29456 4.0e-01
m2a definition_3grp 9.5e-02 0.20515 0.46 0.644 -0.30739 5.0e-01
m2b N -2.2e-04 0.00034 -0.64 0.520 -0.00088 4.4e-04
m2b N_cases -4.4e-04 0.00093 -0.47 0.638 -0.00226 1.4e-03
m2b prevalence -1.4e-03 0.00784 -0.18 0.857 -0.01678 1.4e-02
m2b definition -7.9e-02 0.20903 -0.38 0.705 -0.48890 3.3e-01
m2b definition_3grp -8.1e-02 0.24473 -0.33 0.739 -0.56115 4.0e-01
m2c N -2.0e-04 0.00015 -1.35 0.176 -0.00050 9.1e-05
m2c N_cases -7.7e-05 0.00044 -0.18 0.861 -0.00094 7.9e-04
m2c prevalence 5.9e-03 0.00328 1.79 0.074 -0.00056 1.2e-02
m2c definition -3.3e-02 0.09279 -0.35 0.723 -0.21473 1.5e-01
m2c definition_3grp -1.2e-01 0.09152 -1.26 0.208 -0.29451 6.4e-02

model a

Table 7: early-onset AD, model a pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.72 4.53
ALSPAC DCHS 0.37 15.56
CHS DCHS 0.31 11.04
ALSPAC EDEN 0.74 8.47
CHS EDEN 0.49 12.99
DCHS EDEN 0.38 24.03
ALSPAC GENR 0.56 2.29
CHS GENR 0.38 2.23
DCHS GENR 0.19 13.27
EDEN GENR 0.48 10.76
ALSPAC GOYA 0.56 17.79
CHS GOYA 0.36 13.27
DCHS GOYA 0.34 2.23
EDEN GOYA 0.42 26.26
GENR GOYA 0.62 15.50
ALSPAC INMA 0.80 1.49
CHS INMA 0.50 3.04
DCHS INMA 0.49 14.08
EDEN INMA 0.65 9.95
GENR INMA 0.61 0.80
GOYA INMA 0.76 16.31
ALSPAC IOW 0.71 10.49
CHS IOW 0.60 5.97
DCHS IOW 0.58 5.07
EDEN IOW 0.62 18.96
GENR IOW 0.37 8.20
GOYA IOW 0.23 7.30
INMA IOW 0.62 9.01
ALSPAC IOW-F2 0.64 8.96
CHS IOW-F2 0.48 4.44
DCHS IOW-F2 0.24 6.60
EDEN IOW-F2 0.41 17.43
GENR IOW-F2 0.33 6.67
GOYA IOW-F2 0.36 8.83
INMA IOW-F2 0.53 7.48
IOW IOW-F2 0.57 1.53
ALSPAC Meta 0.95 NA
CHS Meta 0.71 NA
DCHS Meta 0.49 NA
EDEN Meta 0.77 NA
GENR Meta 0.60 NA
GOYA Meta 0.64 NA
INMA Meta 0.91 NA
IOW Meta 0.75 NA
IOW-F2 Meta 0.63 NA
ALSPAC MoBa1 0.64 11.02
CHS MoBa1 0.41 15.54
DCHS MoBa1 0.45 26.58
EDEN MoBa1 0.67 2.55
GENR MoBa1 0.49 13.31
GOYA MoBa1 0.63 28.81
INMA MoBa1 0.86 12.50
IOW MoBa1 0.55 21.51
IOW-F2 MoBa1 0.44 19.98
Meta MoBa1 0.82 NA
ALSPAC MoBa2 0.57 29.07
CHS MoBa2 0.47 33.59
DCHS MoBa2 0.71 44.63
EDEN MoBa2 0.65 20.60
GENR MoBa2 0.13 31.36
GOYA MoBa2 0.31 46.86
INMA MoBa2 0.49 30.55
IOW MoBa2 0.65 39.56
IOW-F2 MoBa2 0.53 38.03
Meta MoBa2 0.63 NA
MoBa1 MoBa2 0.54 18.05
ALSPAC NEST_B 0.73 0.13
CHS NEST_B 0.59 4.40
DCHS NEST_B 0.42 15.44
EDEN NEST_B 0.61 8.59
GENR NEST_B 0.50 2.16
GOYA NEST_B 0.70 17.67
INMA NEST_B 0.83 1.36
IOW NEST_B 0.52 10.37
IOW-F2 NEST_B 0.37 8.84
Meta NEST_B 0.84 NA
MoBa1 NEST_B 0.82 11.14
MoBa2 NEST_B 0.43 29.19
ALSPAC UKIDS 0.91 10.25
CHS UKIDS 0.63 14.77
DCHS UKIDS 0.21 25.81
EDEN UKIDS 0.55 1.78
GENR UKIDS 0.48 12.54
GOYA UKIDS 0.59 28.04
INMA UKIDS 0.79 11.73
IOW UKIDS 0.53 20.74
IOW-F2 UKIDS 0.62 19.21
Meta UKIDS 0.88 NA
MoBa1 UKIDS 0.65 0.77
MoBa2 UKIDS 0.37 18.82
NEST_B UKIDS 0.72 10.37
ALSPAC VITO 0.60 3.91
CHS VITO 0.38 0.62
DCHS VITO 0.34 11.65
EDEN VITO 0.45 12.37
GENR VITO 0.10 1.62
GOYA VITO 0.43 13.89
INMA VITO 0.60 2.42
IOW VITO 0.49 6.59
IOW-F2 VITO 0.33 5.05
Meta VITO 0.61 NA
MoBa1 VITO 0.55 14.93
MoBa2 VITO 0.53 32.98
NEST_B VITO 0.54 3.78
UKIDS VITO 0.51 14.16
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of early-onset eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of early-onset eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 10: Correlation between the effect estimates of early-onset eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 11: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 12: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model b

Table 8: early-onset AD, model b pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.464 4.53
ALSPAC DCHS -0.011 15.90
CHS DCHS 0.369 11.38
ALSPAC EDEN 0.514 8.47
CHS EDEN 0.581 12.99
DCHS EDEN 0.323 24.37
ALSPAC GENR 0.673 2.29
CHS GENR 0.359 2.23
DCHS GENR -0.035 13.61
EDEN GENR 0.563 10.76
ALSPAC GOYA 0.063 17.79
CHS GOYA 0.344 13.27
DCHS GOYA 0.524 1.89
EDEN GOYA 0.039 26.26
GENR GOYA 0.024 15.50
ALSPAC INMA 0.775 1.79
CHS INMA 0.528 2.73
DCHS INMA 0.384 14.11
EDEN INMA 0.631 10.26
GENR INMA 0.658 0.50
GOYA INMA 0.211 16.00
ALSPAC IOW 0.406 10.28
CHS IOW 0.234 5.75
DCHS IOW -0.034 5.63
EDEN IOW 0.169 18.74
GENR IOW 0.189 7.98
GOYA IOW -0.099 7.52
INMA IOW 0.489 8.48
ALSPAC IOW-F2 0.571 8.88
CHS IOW-F2 0.057 4.36
DCHS IOW-F2 -0.041 7.02
EDEN IOW-F2 0.102 17.35
GENR IOW-F2 0.355 6.59
GOYA IOW-F2 -0.037 8.91
INMA IOW-F2 0.493 7.09
IOW IOW-F2 0.438 1.39
ALSPAC Meta 0.955 NA
CHS Meta 0.552 NA
DCHS Meta 0.176 NA
EDEN Meta 0.566 NA
GENR Meta 0.665 NA
GOYA Meta 0.139 NA
INMA Meta 0.882 NA
IOW Meta 0.536 NA
IOW-F2 Meta 0.580 NA
ALSPAC MoBa1 0.791 11.10
CHS MoBa1 0.423 15.63
DCHS MoBa1 0.011 27.00
EDEN MoBa1 0.273 2.63
GENR MoBa1 0.529 13.39
GOYA MoBa1 -0.008 28.89
INMA MoBa1 0.579 12.89
IOW MoBa1 0.415 21.38
IOW-F2 MoBa1 0.424 19.98
Meta MoBa1 0.797 NA
ALSPAC MoBa2 0.684 29.07
CHS MoBa2 0.327 33.59
DCHS MoBa2 0.082 44.97
EDEN MoBa2 0.298 20.60
GENR MoBa2 0.319 31.36
GOYA MoBa2 -0.200 46.86
INMA MoBa2 0.503 30.86
IOW MoBa2 0.523 39.34
IOW-F2 MoBa2 0.439 37.95
Meta MoBa2 0.702 NA
MoBa1 MoBa2 0.708 17.97
ALSPAC NEST_B 0.547 0.13
CHS NEST_B 0.353 4.40
DCHS NEST_B 0.281 15.78
EDEN NEST_B 0.208 8.59
GENR NEST_B 0.173 2.16
GOYA NEST_B 0.289 17.67
INMA NEST_B 0.649 1.66
IOW NEST_B 0.527 10.15
IOW-F2 NEST_B 0.443 8.75
Meta NEST_B 0.642 NA
MoBa1 NEST_B 0.448 11.23
MoBa2 NEST_B 0.350 29.19
ALSPAC UKIDS 0.871 10.25
CHS UKIDS 0.540 14.77
DCHS UKIDS 0.077 26.15
EDEN UKIDS 0.447 1.78
GENR UKIDS 0.523 12.54
GOYA UKIDS 0.202 28.04
INMA UKIDS 0.809 12.04
IOW UKIDS 0.471 20.52
IOW-F2 UKIDS 0.508 19.13
Meta UKIDS 0.899 NA
MoBa1 UKIDS 0.637 0.85
MoBa2 UKIDS 0.539 18.82
NEST_B UKIDS 0.601 10.37
ALSPAC VITO 0.479 3.91
CHS VITO 0.038 0.62
DCHS VITO -0.470 11.99
EDEN VITO 0.197 12.37
GENR VITO 0.252 1.62
GOYA VITO -0.291 13.89
INMA VITO 0.048 2.12
IOW VITO -0.093 6.37
IOW-F2 VITO 0.115 4.97
Meta VITO 0.299 NA
MoBa1 VITO 0.478 15.01
MoBa2 VITO 0.283 32.98
NEST_B VITO -0.053 3.78
UKIDS VITO 0.208 14.16
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of early-onset eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of early-onset eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 13: Correlation between the effect estimates of early-onset eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 14: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 15: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model c

Table 9: early-onset AD, model c pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.5173 4.53
ALSPAC DCHS 0.0607 15.90
CHS DCHS 0.4605 11.38
ALSPAC EDEN 0.5178 8.47
CHS EDEN 0.5985 12.99
DCHS EDEN 0.2443 24.37
ALSPAC GENR 0.5021 2.29
CHS GENR 0.6697 2.23
DCHS GENR 0.1961 13.61
EDEN GENR 0.6859 10.76
ALSPAC GOYA 0.3591 17.79
CHS GOYA 0.2074 13.27
DCHS GOYA 0.0931 1.89
EDEN GOYA -0.0086 26.26
GENR GOYA 0.1570 15.50
ALSPAC INMA 0.5337 1.79
CHS INMA 0.7530 2.73
DCHS INMA 0.3702 14.11
EDEN INMA 0.6319 10.26
GENR INMA 0.8378 0.50
GOYA INMA 0.0266 16.00
ALSPAC IOW 0.6093 10.28
CHS IOW 0.5160 5.75
DCHS IOW 0.0437 5.63
EDEN IOW 0.5477 18.74
GENR IOW 0.5974 7.98
GOYA IOW 0.1333 7.52
INMA IOW 0.7487 8.48
ALSPAC IOW-F2 0.2976 8.88
CHS IOW-F2 0.4564 4.36
DCHS IOW-F2 0.1521 7.02
EDEN IOW-F2 0.5387 17.35
GENR IOW-F2 0.5272 6.59
GOYA IOW-F2 -0.2518 8.91
INMA IOW-F2 0.5156 7.09
IOW IOW-F2 0.3806 1.39
ALSPAC Meta 0.8016 NA
CHS Meta 0.7673 NA
DCHS Meta 0.2351 NA
EDEN Meta 0.7356 NA
GENR Meta 0.8446 NA
GOYA Meta 0.2361 NA
INMA Meta 0.8722 NA
IOW Meta 0.7661 NA
IOW-F2 Meta 0.5236 NA
ALSPAC MoBa1 0.5626 11.10
CHS MoBa1 0.6755 15.63
DCHS MoBa1 0.3529 27.00
EDEN MoBa1 0.4900 2.63
GENR MoBa1 0.6470 13.39
GOYA MoBa1 0.3150 28.89
INMA MoBa1 0.7054 12.89
IOW MoBa1 0.6422 21.38
IOW-F2 MoBa1 0.2879 19.98
Meta MoBa1 0.8110 NA
ALSPAC MoBa2 0.5196 29.07
CHS MoBa2 0.2449 33.59
DCHS MoBa2 0.0737 44.97
EDEN MoBa2 0.3981 20.60
GENR MoBa2 0.2523 31.36
GOYA MoBa2 0.2560 46.86
INMA MoBa2 0.2476 30.86
IOW MoBa2 0.5140 39.34
IOW-F2 MoBa2 0.2832 37.95
Meta MoBa2 0.4610 NA
MoBa1 MoBa2 0.2624 17.97
ALSPAC NEST_B 0.5412 0.13
CHS NEST_B 0.5905 4.40
DCHS NEST_B 0.2607 15.78
EDEN NEST_B 0.4961 8.59
GENR NEST_B 0.5697 2.16
GOYA NEST_B 0.1642 17.67
INMA NEST_B 0.6139 1.66
IOW NEST_B 0.3710 10.15
IOW-F2 NEST_B 0.1080 8.75
Meta NEST_B 0.6942 NA
MoBa1 NEST_B 0.5343 11.23
MoBa2 NEST_B 0.1217 29.19
ALSPAC UKIDS 0.7279 10.25
CHS UKIDS 0.7402 14.77
DCHS UKIDS 0.1058 26.15
EDEN UKIDS 0.5831 1.78
GENR UKIDS 0.7375 12.54
GOYA UKIDS 0.3411 28.04
INMA UKIDS 0.7111 12.04
IOW UKIDS 0.5899 20.52
IOW-F2 UKIDS 0.4299 19.13
Meta UKIDS 0.8861 NA
MoBa1 UKIDS 0.6560 0.85
MoBa2 UKIDS 0.2739 18.82
NEST_B UKIDS 0.7121 10.37
ALSPAC VITO 0.3307 3.91
CHS VITO -0.2391 0.62
DCHS VITO -0.4712 11.99
EDEN VITO 0.1425 12.37
GENR VITO 0.0423 1.62
GOYA VITO 0.1611 13.89
INMA VITO -0.0088 2.12
IOW VITO 0.3068 6.37
IOW-F2 VITO 0.0417 4.97
Meta VITO 0.1880 NA
MoBa1 VITO 0.1394 15.01
MoBa2 VITO 0.2746 32.98
NEST_B VITO -0.0166 3.78
UKIDS VITO 0.1430 14.16
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of early-onset eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of early-onset eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 16: Correlation between the effect estimates of early-onset eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 17: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 18: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Persistent AD

Sample sizes + prevalence

cohort N N-cases N-controls prevalence
ALSPAC 662 125 537 18.9
CHS 197 35 162 17.8
EDEN 126 18 108 14.3
GENR 415 39 376 9.4
MoBa1 519 87 432 16.8
MoBa2 237 61 176 25.7
VITO 66 11 55 16.7
GOYA 517 33 484 6.4
IOW 462 25 437 5.4
Total 3,201 434 2,767 13.6

Meta-regression results

Table 10: Assessing the impact of prevalence and samplesize on the influence a study has in the meta-analysis results
model variable beta se zval pval lower-ci upper-ci
m3a N -0.00024 0.00027 -0.89 0.37 -0.00077 0.00029
m3a N_cases -0.00027 0.00158 -0.17 0.86 -0.00336 0.00282
m3a prevalence 0.01235 0.00762 1.62 0.11 -0.00258 0.02728
m3a definition -0.05261 0.17375 -0.30 0.76 -0.39316 0.28794
m3a definition_3grp -0.08742 0.19265 -0.45 0.65 -0.46501 0.29017
m3b N -0.00042 0.00024 -1.75 0.08 -0.00089 0.00005
m3b N_cases -0.00140 0.00150 -0.93 0.35 -0.00434 0.00154
m3b prevalence 0.00991 0.00817 1.21 0.22 -0.00609 0.02592
m3b definition -0.15004 0.16706 -0.90 0.37 -0.47747 0.17740
m3b definition_3grp -0.22939 0.16026 -1.43 0.15 -0.54350 0.08471
m3c N -0.00021 0.00033 -0.64 0.52 -0.00086 0.00044
m3c N_cases -0.00133 0.00166 -0.80 0.42 -0.00458 0.00192
m3c prevalence -0.00349 0.00863 -0.40 0.69 -0.02040 0.01342
m3c definition -0.08385 0.20665 -0.41 0.68 -0.48887 0.32118
m3c definition_3grp -0.16820 0.20856 -0.81 0.42 -0.57697 0.24057

model a

Table 11: persistent AD, model a pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.589 1.116
ALSPAC EDEN 0.267 4.596
CHS EDEN -0.104 3.481
ALSPAC GENR 0.788 9.485
CHS GENR 0.415 8.369
EDEN GENR 0.115 4.888
ALSPAC GOYA 0.566 12.499
CHS GOYA 0.538 11.384
EDEN GOYA 0.151 7.903
GENR GOYA 0.457 3.015
ALSPAC IOW 0.536 13.471
CHS IOW 0.276 12.355
EDEN IOW 0.048 8.874
GENR IOW 0.533 3.986
GOYA IOW 0.190 0.972
ALSPAC Meta 0.932 NA
CHS Meta 0.643 NA
EDEN Meta 0.157 NA
GENR Meta 0.869 NA
GOYA Meta 0.605 NA
IOW Meta 0.642 NA
ALSPAC MoBa1 0.586 2.119
CHS MoBa1 0.561 1.003
EDEN MoBa1 -0.166 2.477
GENR MoBa1 0.607 7.365
GOYA MoBa1 0.241 10.380
IOW MoBa1 0.532 11.352
Meta MoBa1 0.763 NA
ALSPAC MoBa2 0.528 6.856
CHS MoBa2 0.410 7.972
EDEN MoBa2 -0.040 11.453
GENR MoBa2 0.541 16.341
GOYA MoBa2 0.610 19.355
IOW MoBa2 0.367 20.327
Meta MoBa2 0.658 NA
MoBa1 MoBa2 0.427 8.975
ALSPAC VITO 0.396 2.216
CHS VITO 0.170 1.100
EDEN VITO 0.157 2.381
GENR VITO 0.415 7.269
GOYA VITO 0.176 10.284
IOW VITO 0.208 11.255
Meta VITO 0.467 NA
MoBa1 VITO 0.214 0.096
MoBa2 VITO 0.577 9.072
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of persistent eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of persistent eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 19: Correlation between the effect estimates of persistent eczema EWAS, model a, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 20: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 21: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model b

Table 12: persistent AD, model b pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.5602 1.12
ALSPAC EDEN 0.3047 4.60
CHS EDEN -0.0056 3.48
ALSPAC GENR 0.7850 9.48
CHS GENR 0.4789 8.37
EDEN GENR 0.2270 4.89
ALSPAC GOYA 0.6058 12.50
CHS GOYA 0.4684 11.38
EDEN GOYA 0.3189 7.90
GENR GOYA 0.4308 3.01
ALSPAC IOW 0.4910 13.57
CHS IOW 0.3674 12.46
EDEN IOW 0.3370 8.98
GENR IOW 0.5328 4.09
GOYA IOW 0.2302 1.07
ALSPAC Meta 0.9274 NA
CHS Meta 0.6078 NA
EDEN Meta 0.2931 NA
GENR Meta 0.8478 NA
GOYA Meta 0.6121 NA
IOW Meta 0.6360 NA
ALSPAC MoBa1 0.5652 2.05
CHS MoBa1 0.5013 0.94
EDEN MoBa1 -0.0313 2.54
GENR MoBa1 0.5309 7.43
GOYA MoBa1 0.2461 10.44
IOW MoBa1 0.4589 11.52
Meta MoBa1 0.7507 NA
ALSPAC MoBa2 0.4231 6.86
CHS MoBa2 0.0625 7.97
EDEN MoBa2 0.0586 11.45
GENR MoBa2 0.3304 16.34
GOYA MoBa2 0.4391 19.36
IOW MoBa2 0.3887 20.43
Meta MoBa2 0.5317 NA
MoBa1 MoBa2 0.3945 8.91
ALSPAC VITO 0.3447 2.22
CHS VITO -0.0691 1.10
EDEN VITO -0.0578 2.38
GENR VITO 0.2397 7.27
GOYA VITO 0.0946 10.28
IOW VITO 0.0071 11.36
Meta VITO 0.3710 NA
MoBa1 VITO 0.2426 0.16
MoBa2 VITO 0.4350 9.07
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of persistent eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of persistent eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 22: Correlation between the effect estimates of persistent eczema EWAS, model b, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 23: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 24: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

model c

Table 13: persistent AD, model c pairwise cohort comparsions
cohort1 cohort2 effect-cor prev-diff
ALSPAC CHS 0.789 18.88
ALSPAC EDEN 0.543 4.60
CHS EDEN 0.325 14.29
ALSPAC GENR 0.801 9.48
CHS GENR 0.590 9.40
EDEN GENR 0.479 4.89
ALSPAC GOYA 0.767 12.50
CHS GOYA 0.611 6.38
EDEN GOYA 0.316 7.90
GENR GOYA 0.552 3.01
ALSPAC IOW 0.583 13.57
CHS IOW 0.509 5.31
EDEN IOW 0.757 8.98
GENR IOW 0.509 4.09
GOYA IOW 0.239 1.07
ALSPAC Meta 0.975 NA
CHS Meta 0.809 NA
EDEN Meta 0.617 NA
GENR Meta 0.844 NA
GOYA Meta 0.729 NA
IOW Meta 0.703 NA
ALSPAC MoBa1 0.861 2.05
CHS MoBa1 0.796 16.83
EDEN MoBa1 0.548 2.54
GENR MoBa1 0.732 7.43
GOYA MoBa1 0.538 10.44
IOW MoBa1 0.671 11.52
Meta MoBa1 0.910 NA
ALSPAC MoBa2 0.724 6.86
CHS MoBa2 0.747 25.74
EDEN MoBa2 0.355 11.45
GENR MoBa2 0.528 16.34
GOYA MoBa2 0.669 19.36
IOW MoBa2 0.434 20.43
Meta MoBa2 0.758 NA
MoBa1 MoBa2 0.722 8.91
ALSPAC VITO 0.274 2.22
CHS VITO 0.137 16.67
EDEN VITO 0.050 2.38
GENR VITO 0.343 7.27
GOYA VITO 0.033 10.28
IOW VITO 0.184 11.36
Meta VITO 0.263 NA
MoBa1 VITO 0.080 0.16
MoBa2 VITO 0.174 9.07
effect-cor = correlation between effect estimates of the top 30 CpGs, prev-diff = the estimated prevalence difference between cohorts. For more information see the Summary of project + methods section.
Correlation between the effect estimates of persistent eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.Correlation between the effect estimates of persistent eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

Figure 25: Correlation between the effect estimates of persistent eczema EWAS, model c, across different cohorts. The left-hand-side is a heatmap showing correlations between effect estimates of the top 30 CpG sites (top meaning those with the lowest P value from the meta-analysis). The right-hand-side is a heatmap showing correlations between effect estimates of all overlapping CpG sites.

M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Figure 26: M-statistics, prevalence and effect size These plots show calculated M-statistics, a measure of heterogeneity between studies using the top 30 CpG sites, for each study and their association with prevalence and effect size. A: the distribution of m-statistics, B: the association between M-statistics and prevalence, C: the association between M-statistics and effect size.

Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.

Figure 27: Comparison of effect sizes between cohorts The bars represent the 95% confidence intervals for the meta-analysis results.